Eurostar’s AI Chatbot Vulnerabilities Exposed by Security Researchers Security researchers at Pen Test Partners uncovered critical flaws in Eurostar’s AI-powered customer support chatbot, which could have enabled malicious exploitation. The vulnerabilities included weak message validation, allowing attackers to manipulate older prompts to execute unauthorized actions—such as extracting system details or potentially exfiltrating data. Additional weaknesses involved unverified conversation IDs and an HTML injection flaw, which permitted JavaScript execution within the chat interface. Despite the risks, Eurostar confirmed that no customer data was compromised, as the chatbot lacked access to sensitive databases or login-protected information. The company stated that all vulnerabilities have since been mitigated, emphasizing that the chatbot operated in isolation from critical systems. The discovery highlights broader concerns around rapid AI adoption in enterprises, where misconfigurations and non-human identities are expanding cloud attack surfaces. While Pen Test Partners did not exploit the flaws to access user data, they warned that similar design weaknesses could pose greater risks as chatbot functionality evolves. The incident underscores the need for rigorous security testing in AI-driven tools, particularly as businesses integrate them into customer-facing operations.

Eurostar asked all of its customers to reset their passwords after detecting an “unauthorised attempt” to hack into its systems and access their accounts. The company identified an attempt to access eurostar.com accounts using users’ email and passwords between the 15 and 19 of October, after which it asked its customers to reset the passwords.